Stress, which commonly occurs among university students, is often not adequately addressed because of limited access to mental health services. This emphasizes the need for easily available and scalable treatments. Artificial Intelligence (AI)-driven mental health chatbots provide a helpful option by giving students customized and instantaneous advice to effectively manage stress. This paper evaluated the efficacy of an AI-powered chatbot using the Trier Social Stress Test (TSST), a validated method for inducing acute psychological stress. Thirty-four participants completed the TSST, which consists of four stages: baseline, anticipation, post-task, and recovery. After each stage, participants completed psychological stress assessments (Visual Analog Scale and State Trait Anxiety Inventory state scale) and physiological assessments (blood pressure and heart rate). During anticipation, the experimental group interacted with the chatbot, while the control group remained silently at rest. Stress was induced using a mock job interview and mental arithmetic task. Although no statistically significant group-by-interval interaction effects were observed, the chatbot group showed modest reductions in anxiety and altered autonomic responses during anticipation. These findings suggest AI-based tools may support stress management, even if their effectiveness remains inconclusive. Future studies should use larger, more diverse samples and include physiological measures such as heart rate variability and galvanic skin response to improve accuracy and allow more personalized insights. Despite certain limitations, this study enriches the growing literature on the use of AI in mental health care. It emphasizes the need for more research on how user-adaptable technologies can contribute to conventional mental health treatment, especially in educational environments where efficient stress management is crucial.